【摘要】 关系作为一类重要知识，在管理决策中扮演关键角色，通过创新性地设计智能方法从大数据高阶空间中发现新颖关系是管理科学领域的关注焦点。本文对数智化管理决策中的“新颖关系发现”问题进行系统性探讨，基于对相关概念和文献的梳理界定分析了其科学内涵、表示方式和管理意义，以及数据、知识和映射视角下的要素特征。进而，对若干前沿探索进行凝练，包括融合消费者行为和体验的产品间关系、由兴趣属性间接传导的品牌间关系、协同电商和社交领域的嵌入间关系等，给出新颖关系发现的问题建模、知识表示和求解策略，并对未来研究方向进行展望。特别地，在复杂管理场景和科学决策需求下，数据更多具有宏微粒度、内外来源、富多模态、交叉时变的特征，且知识更多呈现间接、隐性、动态、跨域的形式，而智能方法创新则需要基于问题情境设计序列学习、协同寻优、向量嵌入、潜隐变量等技术要素，以充分融合和表征大数据并有效构建高阶空间，通过深层语义建模来发现新颖关系。本文工作有助于学界和业界获得对于“新颖关系发现”问题的深入洞察，为管理科学领域大数据分析方法创新和新知贡献提供启迪，同时对数智赋能的商业模式创新以及科学政策决策提供支撑和参考价值。更多还原

【Abstract】 Relations,as an important type of knowledge,play crucial roles in managerial decision making,and the innovative design of intelligent approaches for novel relation discovery from high-dimensional spaces of big data is one of the focal points in management science research.This article systematically examines the problem of novel relation discovery in data-intelligence-driven managerial decision making.Based on a comprehensive review of related concepts and literature,it elaborates on the problem’s conceptualization,formulation,and managerial implications,along with the characteristics from the viewpoints of data,knowledge,and mapping.Furthermore,the study highlights several exploratory advances.These include studies on product relations integrated with consumer behavior and experience,brand relations indirectly connected through attributes of interest,and embedding relations derived from collaborating e-commerce and social domains.Each is discussed in terms of problem modeling,knowledge forms,and solving strategies,followed by a prospective outlook.Specifically,with complex managerial scenarios and scientific decision-making needs,data is characterized by macro-and micro-level granularity,internal and external sources,rich multimodality,and cross-temporal variance.Meanwhile,relational knowledge tends to manifest in indirect,implicit,dynamic,and cross-domain forms.Consequently,innovations in intelligent methods call for the design of technical components such as sequence learning,collaborative optimization,vector embedding,and latent variables tailored to specific problem contexts.These approaches should fully integrate and formulate big data to effectively construct high-order spaces,thereby enabling the discovery of novel relations through modeling underlying semantics.This article offers in-depth insights into the problem of novel relation discovery for both academia and practitioners.It also sheds light on methodological innovations and intellectual contributions within big data analytics.Furthermore,it provides a valuable reference for data-intelligence empowered business models and scientific policy decisionmaking.更多还原